> ## Documentation Index
> Fetch the complete documentation index at: https://docs.retrofix.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Browser Automation

> Automate web-based tasks with hosted browser agents

Browser automation allows you to spin up a hosted browser in the cloud and have it take actions using data from previous workflow steps or send data to subsequent steps.

***

## What is Browser Automation?

Use Browser Agents as a step in your workflow to automate web-based tasks directly. This is one of the most powerful tools in RetroFix for handling applications without native integrations.

***

## Use Cases

### Integrate with Unsupported Applications

Set up a browser workflow to take actions on applications that RetroFix doesn't have native integration for—just like you would on your own computer.

**Example:** Fill out a form, navigate through a multi-step process, or interact with a custom web application.

### Scrape Dynamic Data

Use browser automation to scrape websites when you need the most up-to-date information that may load dynamically.

**Example:** Every Sunday, fetch a list of competitor URLs from Google Sheets, spin up a browser agent for each one, scrape their pricing, compare against the previous week, and send a Slack message alert if anything changed.

***

## Browser Actions

The browser configuration panel has a list of browser actions on the left and a preview on the right. There are 4 main action types:

### Navigate

Navigate to a specific URL.

**Options:**

* Basic navigation with URL input
* Click the gear icon to control navigation timeout in case the website takes too long to load

**Example:** [https://www.retrofix.ai](https://www.retrofix.ai)

***

### Act

The main step to control the browser. Takes in an English instruction and supports variables (type `@` to see them).

#### Two modes:

<Tabs>
  <Tab title="Action Mode">
    AI generates the action once, then runs it deterministically with self-healing.

    **Best for:** Specific, repeatable instructions

    **Example:** "click the button that says Integrations"

    **Advantages:**

    * Faster execution
    * More reliable and predictable
    * Doesn't need to think through each action
  </Tab>

  <Tab title="Agent Mode">
    Choose from multiple agents (RetroFix, OpenAI, Anthropic, Gemini). Agents handle complex instructions but think through each action.

    **Best for:** Complex, uncertain, or multi-step instructions

    **Example:** "Find the pricing page and extract all the pricing tiers"

    **Advantages:**

    * Handles more complex scenarios
    * Better at adapting to page changes
    * More flexible instruction format

    **Trade-off:** Slightly slower since the agent reasons through each action
  </Tab>
</Tabs>

***

### Extract

Extract structured information from a page given a schema.

**Use cases:**

* Extract pricing information from competitor sites
* Pull product details from listings
* Scrape structured data like tables or product specifications

#### Two modes:

<Tabs>
  <Tab title="Action Mode">
    Extracts with whatever is currently visible on the page. Does not perform additional actions.

    **Best for:** When the data is already visible without interaction
  </Tab>

  <Tab title="Agent Mode">
    Agent will attempt to find the information on the website however it can before extracting. May need to scroll, click, or navigate to find data.

    **Best for:** When data requires navigation or interaction to find
  </Tab>
</Tabs>

***

### Wait

Wait for conditions or time to pass on the browser.

**Options:**

* **Seconds** - Wait for a specific duration
* **Network Idle** - Wait until all network requests complete
* **DOM Loaded** - Wait until DOM is fully loaded
* **Page Load** - Wait for full page load
* **No Network Activity** - Wait until network activity stops

**Example:** After clicking a button, wait for the page to load before extracting data.

***

## Best Practices

* Use **Action mode** for simple, deterministic tasks
* Use **Agent mode** for complex or variable instructions
* Always use **Wait** appropriately after navigation or interactions
* Test browser steps thoroughly before publishing
* Consider timeout values for slower-loading sites